Siwx91x dma driver fixes #88705

smalae · 2025-04-16T10:05:23Z

The following fixes are done in the DMA driver:

Introduced a new variable in the dma_siwx91x_channel_info
structure to provide a clean way to differentiate transfer
directions. This enhancement is utilized to trigger software
requests specifically for memory-to-memory transfers
Addressed an issue where regular/non-scatter-gather DMA
transfers were not explicitly using the primary DMA descriptor
structure. This ensures a smooth regular DMA transfer after
any scatter-gather transfer.
Addressed an issue where alignment of dma desc varaible
of dma0 is corrected to 1024 instead of 512.

jhedberg · 2025-04-17T08:38:18Z

drivers/dma/dma_silabs_siwx91x.c

@@ -37,6 +37,7 @@ struct dma_siwx91x_channel_info {
 	dma_callback_t dma_callback;        /* User callback */
 	void *cb_data;                      /* User callback data */
 	RSI_UDMA_DESC_T *sg_desc_addr_info; /* Scatter-Gather table start address */
+	uint8_t transfer_direction;         /* mem<->mem ot per<->mem */


I think it's worth to consider giving a named type for the enum and just use that everywhere instead of the mix of uint8_t and int. Btw, you don't gain anything with using a smaller type in the above struct due to implicit padding, i.e. the struct size would be exactly the same if you used int. Even if you did need a smaller type, you can also get that with an enum by adding __packed to the definition. I'd also consider adding a third possible value to make the code look more intuitive (e.g. you can return an actual enum value instead of -EINVAL):

enum transfer_dir { TRANSFER_MEM_TO_MEM, TRANSFER_TO_OR_FROM_PER, TRANSFER_DIR_INVALID = -1, };

Perhaps out of scope for this PR, but if you want to make these symbol names more compact, I believe a common convention is to shorten "transfer" to "xfer'.

I've updated it according to the suggestion.

jerome-pouiller · 2025-04-17T09:53:25Z

drivers/dma/dma_silabs_siwx91x.c

@@ -464,8 +478,7 @@ static int siwx91x_dma_start(const struct device *dev, uint32_t channel)
 	}

 	/* Check if the transfer type is memory-memory */
-	if (udma_table[channel].vsUDMAChaConfigData1.srcInc != UDMA_SRC_INC_NONE &&
-	    udma_table[channel].vsUDMAChaConfigData1.dstInc != UDMA_DST_INC_NONE) {
+	if (data->zephyr_channel_info[channel].xfer_direction == TRANSFER_MEM_TO_MEM) {


Just curiosity, in which case the original condition didn't work?

In case of transfers using alternate descriptors, this condition is true for all the transfers and triggers the SW request even for peripheral transfers.

jhedberg · 2025-04-17T17:24:58Z

drivers/dma/dma_silabs_siwx91x.c

@@ -206,13 +208,13 @@ static int siwx91x_sg_chan_config(const struct device *dev, RSI_UDMA_HANDLE_T ud
 	struct dma_siwx91x_data *data = dev->data;
 	RSI_UDMA_DESC_T *sg_desc_base_addr = NULL;
 	uint8_t transfer_type;
-	int ret;
+	int xfer_dir;


enum xfer_dir dir;

jhedberg · 2025-04-17T17:25:35Z

drivers/dma/dma_silabs_siwx91x.c

-	ret = siwx91x_transfer_direction(config->channel_direction);
-	if (ret < 0) {
+	xfer_dir = siwx91x_transfer_direction(config->channel_direction);
+	if (xfer_dir < 0) {


if (dir == TRANSFER_DIR_INVALID) {

jhedberg · 2025-04-17T17:26:49Z

drivers/dma/dma_silabs_siwx91x.c

 		return -EINVAL;
 	}
-	transfer_type = ret ? UDMA_MODE_PER_SCATTER_GATHER : UDMA_MODE_MEM_SCATTER_GATHER;
+	transfer_type = xfer_dir ? UDMA_MODE_PER_SCATTER_GATHER : UDMA_MODE_MEM_SCATTER_GATHER;


if (dir == TRANSFER_TO_OR_FROM_PER) { transfer_type = UDMA_MODE_PER_SCATTER_GATHER; } else { transfer_type = UDMA_MODE_MEM_SCATTER_GATHER; }

(xfer_dir is not a boolean so don't treat it like that)

jerome-pouiller · 2025-04-18T07:08:43Z

drivers/dma/dma_silabs_siwx91x.c

@@ -28,15 +28,17 @@

 LOG_MODULE_REGISTER(si91x_dma, CONFIG_DMA_LOG_LEVEL);

-enum {
+enum xfer_dir_t {


Zephyr tends to apply the Coding Style of the Linux kernel. So, _t is only used for typedefs. In addition, we try to avoid typedefs and use plain enum xx or struct yy instead.

smalae · 2025-04-21T07:11:03Z

@jerome-pouiller @jhedberg . I have added one more commit to this PR, which corrects the alignment of sram_desc_addr from 512 to 1024. This issue was identified during SPI testing with the latest DMA changes introduced by the memory_partition PR. Kindly review.

drivers/dma/dma_silabs_siwx91x.c

jhedberg · 2025-04-22T06:44:33Z

drivers/dma/dma_silabs_siwx91x.c

 };

 struct dma_siwx91x_channel_info {
 	dma_callback_t dma_callback;        /* User callback */
 	void *cb_data;                      /* User callback data */
 	RSI_UDMA_DESC_T *sg_desc_addr_info; /* Scatter-Gather table start address */
+	int xfer_direction;                 /* mem<->mem ot per<->mem */


One more int -> enum dma_xfer_dir here

Introduced a new variable in the `dma_siwx91x_channel_info` structure to provide a clean way to differentiate transfer directions. This enhancement is utilized to trigger software requests specifically for memory-to-memory transfers Signed-off-by: Sai Santhosh Malae <[email protected]>

Addressed an issue where regular/non-scatter-gather DMA transfers were not explicitly using the primary DMA descriptor structure. This ensures a smooth regular DMA transfer after any scatter gather transfer. Signed-off-by: Sai Santhosh Malae <[email protected]>

Addressed an issue where alignment of dma desc varaible of dma0 is corrected to 1024 instead of 512. Signed-off-by: Sai Santhosh Malae <[email protected]>

github-actions bot added area: DMA Direct Memory Access platform: Silabs Silicon Labs labels Apr 16, 2025

github-actions bot requested review from asmellby, jerome-pouiller, jhedberg, Martinhoff-maker and teburd April 16, 2025 10:06

github-actions bot assigned jhedberg and asmellby Apr 16, 2025

jhedberg reviewed Apr 17, 2025

View reviewed changes

smalae force-pushed the siwx91x_dma_driver_fixes branch from d7dc92b to 697b353 Compare April 17, 2025 09:03

smalae requested a review from jhedberg April 17, 2025 09:06

jerome-pouiller previously approved these changes Apr 17, 2025

View reviewed changes

jerome-pouiller reviewed Apr 17, 2025

View reviewed changes

jhedberg reviewed Apr 17, 2025

View reviewed changes

smalae dismissed jerome-pouiller’s stale review via b59f0bf April 18, 2025 06:14

smalae force-pushed the siwx91x_dma_driver_fixes branch from 697b353 to b59f0bf Compare April 18, 2025 06:14

smalae requested review from jhedberg and jerome-pouiller April 18, 2025 06:16

jerome-pouiller previously approved these changes Apr 18, 2025

View reviewed changes

jerome-pouiller reviewed Apr 18, 2025

View reviewed changes

smalae dismissed jerome-pouiller’s stale review via 57a798f April 21, 2025 07:07

smalae force-pushed the siwx91x_dma_driver_fixes branch from b59f0bf to 57a798f Compare April 21, 2025 07:07

smalae requested a review from jerome-pouiller April 21, 2025 07:08

jhedberg previously approved these changes Apr 22, 2025

View reviewed changes

jhedberg reviewed Apr 22, 2025

View reviewed changes

drivers/dma/dma_silabs_siwx91x.c Show resolved Hide resolved

smalae dismissed jhedberg’s stale review via e773c95 April 22, 2025 06:21

smalae force-pushed the siwx91x_dma_driver_fixes branch from 57a798f to e773c95 Compare April 22, 2025 06:21

smalae requested a review from jhedberg April 22, 2025 06:22

jhedberg previously approved these changes Apr 22, 2025

View reviewed changes

jhedberg reviewed Apr 22, 2025

View reviewed changes

smalae added 3 commits April 22, 2025 14:25

drivers: dma: siwx91x: SRAM desc alignment bug fix

e08a1ad

Addressed an issue where alignment of dma desc varaible of dma0 is corrected to 1024 instead of 512. Signed-off-by: Sai Santhosh Malae <[email protected]>

smalae dismissed jhedberg’s stale review via e08a1ad April 22, 2025 08:59

smalae force-pushed the siwx91x_dma_driver_fixes branch from e773c95 to e08a1ad Compare April 22, 2025 08:59

smalae requested a review from jhedberg April 22, 2025 09:00

jhedberg approved these changes Apr 22, 2025

View reviewed changes

jerome-pouiller approved these changes Apr 22, 2025

View reviewed changes

kartben merged commit 0d547d4 into zephyrproject-rtos:main Apr 22, 2025
24 checks passed

jhedberg mentioned this pull request Apr 24, 2025

Contributor Nomination: Sai Santhosh Malae #89002

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Siwx91x dma driver fixes #88705

Siwx91x dma driver fixes #88705

smalae commented Apr 16, 2025 •

edited

Loading

jhedberg Apr 17, 2025

smalae Apr 17, 2025

jerome-pouiller Apr 17, 2025

smalae Apr 18, 2025

jhedberg Apr 17, 2025

smalae Apr 18, 2025

jhedberg Apr 17, 2025

smalae Apr 18, 2025

jhedberg Apr 17, 2025 •

edited

Loading

smalae Apr 18, 2025

jerome-pouiller Apr 18, 2025

smalae Apr 21, 2025

smalae commented Apr 21, 2025

jhedberg Apr 22, 2025

Siwx91x dma driver fixes #88705

Siwx91x dma driver fixes #88705

Conversation

smalae commented Apr 16, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jhedberg Apr 17, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

smalae commented Apr 21, 2025

Choose a reason for hiding this comment

smalae commented Apr 16, 2025 •

edited

Loading

jhedberg Apr 17, 2025 •

edited

Loading